Diabetes mellitus affects an estimated 422 million people globally [1]. Approximately 90% of these cases are of type 2 diabetes [2,3]. Type 2 diabetes is typically managed through diet, physical activity, and oral medication. However, poor adherence to oral glucose-lowering medications in this population is common. The incidence of nonadherence is estimated to be between 38% and 93% [4-6], depending on the method used to define and measure adherence, and up to 37% of patients stop using oral glucose-lowering medications within 1 year of starting treatment [7]. This is a problem because medication nonadherence is associated with poorer clinical and health outcomes in patients with diabetes [8,9] and results in substantial financial burdens on Western health care systems because of higher rates of morbidity and mortality associated with the condition [10,11].

Effective interventions to increase medication adherence are needed to optimize health outcomes, quality of life, and cost-effective health care [12]. Despite the need, there is a lack of convincing evidence for interventions to improve patient adherence to prescribed medications for common chronic health problems. A Cochrane systematic review of 182 adherence-enhancing trials reported inconsistent effects on adherence, from highly heterogeneous interventions, and the majority of the lowest risk of bias trials (12 of 17) reported no improvements in adherence and clinical outcomes [13]. The interventions were generally multicomponent and complex, delivered by health care professionals and involving one-on-one education and counseling. Such interventions are time- and resource-intensive and, therefore, difficult to deliver to a large number of people or implement in usual practice settings. Similarly, complex interventions aimed at supporting medication adherence in patients with diabetes also report few improvements in adherence or clinical outcomes [14-18].

As of May 2014, 7 billion mobile phone subscriptions were in use worldwide [19]. In the United Kingdom, 65% to 92% of retired households own a mobile phone [20], and in 2017, 73% of adults accessed the internet through a smartphone or mobile device, which is double the rate recorded in 2011 [21]. The United States displays similar trends in increased mobile phone ownership; 77% and 95% of adults own a smartphone or mobile phone, respectively [22]. The increased use of mobile phone and modern technologies has created a new medium for digital health interventions delivered by mobile devices to support self-management of long-term conditions and potential solutions to problems such as medication nonadherence. Currently, the short message service (SMS) text message function of mobile phones offers a possible low-cost and wide-reaching means to deliver behavior change interventions. Brief messages delivered by SMS text message can be used in several ways: to send motivational and social support messages [23], to challenge maladaptive beliefs [24], or provide a cue to action [25]. This is important, as the impact on behaviors, such as medication adherence, is likely to differ depending on the message content.

A recent systematic review of brief messaging interventions for patients with type 2 diabetes concluded that these interventions may have the potential for supporting medication adherence [26]. However, the authors noted a widespread lack of explicit theoretical frameworks underpinning the included trials and called for greater use of theory in designing interventions to address the behavioral mechanisms through which changes to adherence may occur [26].

Behavior change theory can inform interventions by providing insight into factors that influence behavior, which can be targeted for behavioral change. This includes theoretical constructs (ie, those variables or constructs from theories that are targeted by interventions, eg, self-efficacy) and mechanisms underlying specific behavioral strategies (ie, techniques not necessarily linked to a single theory but incorporated in interventions because they predict behavior, eg, modeling). To translate these strategies into a standardized language, a comprehensive taxonomy of 93 behavior change techniques (BCTs) has been developed for use in behavior change interventions [27]. BCTs are discrete intervention components used to facilitate behavior change, such as problem solving and goal setting, and can be thought of as the active ingredients in interventions [27]. With this BCT taxonomy (BCTT), researchers can identify and evaluate specific techniques linked to intervention success, which facilitates the replication and optimization of existing or new interventions [27]. There is evidence that interventions that incorporate a greater number of BCTs tend to have larger effects on behavior than those incorporating a smaller number [28].

Grounding research in theory is crucial to progress understanding of which constructs need to change to, in turn, effectively change behavior. However, it is often unclear how message content has been developed, including what, if any, theory the messages are rooted in or how different types of messages may affect behavior. Without this, it is difficult to establish whether messages targeting constructs are effective or how to replicate these interventions. Toolkits for designing SMS-based interventions exist, but guidelines for developing the core message content are lacking (eg, The Center for Research in Implementation Science and Prevention [29]). An established process for developing brief message content, such as those delivered via SMS text message, is needed to optimize intervention effectiveness and facilitate evaluation of the most successful message types for changing behavior.

This review represents the first stage of a larger program of research to identify effective brief message content for use in interventions to support medication adherence in people with type 2 diabetes. We propose a novel approach to develop theory-based brief message content for use in interventions supporting medication adherence for people with type 2 diabetes. Given the lack of effectiveness of existing interventions [13], there is a need to determine promising novel intervention content for the future. For this reason, this rapid systematic review considered the factors related to adherence in 2 stages. First, we considered the factors found to affect adherence across a broad range of physical health conditions in previous systematic reviews of quantitative research. Second, we considered systematic reviews of qualitative research of adherence, specifically in people with diabetes to better understand the psychosocial context of and influences on adherence in our target population.

The aims of this rapid review were to (1) systematically search for systematic reviews of quantitative studies that focused on medication adherence in physical health conditions and systematic reviews of qualitative studies that focused on adherence in type 2 diabetes, (2) extract data on theoretical constructs and behavioral strategies associated with medication adherence, and (3) map these constructs and strategies to BCTs to generate a list of BCTs that may show promise in improving medication adherence. We propose using these BCTs, to inform intervention targets and content of brief messages in future digitally based adherence-enhancing interventions.

We conducted a rapid systematic review of systematic reviews. Rapid systematic reviews use methods to accelerate and streamline traditional systematic review processes while preserving the quality and rigor of review methods [30]. A rapid review is an appropriate method, given our aim to identify new potential intervention targets, rather than to conduct a comprehensive appraisal of the evidence base.

A systematic search of 2 electronic databases (MEDLINE and PsycINFO) was conducted from database inception to April 10, 2017. Search terms for medication adherence were chosen by adapting Medical Subject Headings (MeSH) and keywords previously used [13]. Search strategies were reviewed by the team. The search terms used in MEDLINE were “medication adherence [MeSH]” OR “medication compliance” OR “patient compliance [MeSH]” AND “review literature as topic [MeSH]” OR “systematic review” OR “meta-analysis” OR “meta-synthesis.” The search terms used in PsycINFO were “treatment compliance [MeSH]” OR “compliance [MeSH]” AND “literature review [MeSH]” OR “meta-analysis [MeSH]” OR “systematic review” OR “meta-synthesis.” EndNote reference software (Clarivate Analytics) was used to organize references.

Separate data extraction sheets were developed, piloted, and refined for quantitative and qualitative data. First, we extracted data from the systematic reviews of quantitative studies on a number of indicators (Multimedia Appendix 1) and the relationship between constructs and strategies with medication adherence.

Second, we extracted qualitative data in the form of the authors’ narrative results sections. Results pertaining only to patients’ experiences of adherence were extracted. The mixed-method reviews, and the data reported in these, were combined with either the pool of quantitative or qualitative reviews depending on whether the findings were more quantitative or qualitative in nature.

There were 2 stages involved in the process of identifying and mapping theoretical constructs and behavioral strategies to BCTs. First, we used the reviews of quantitative research to identify foundational constructs and strategies that were positively related to adherence in a range of physical health conditions. For each of the constructs and strategies identified in the reviews of quantitative studies, a judgment was made about which BCTs in the BCTT may be an applicable technique to change that particular construct or strategy. This process was completed iteratively, reviewing and refining judgments to ensure that all constructs and strategies had been paired with at least one BCT, and any relevant BCTs were captured. The process of mapping involved a degree of brainstorming [31] and idea generation, a recognized early step in intervention mapping, with the aim of generating as many explanations as possible in response to a question [32]. We have adapted this approach, generating answers, in the form of BCTs, to the constructs and strategies (the questions) extracted from the reviews.

Second, we took a diabetes-specific approach and used the reviews of qualitative research to consider the specific context of our target population and to ensure that we did not overlook constructs or strategies that were particularly relevant to adherence in patients with diabetes. In this stage, a different method was used to identify constructs and strategies in the qualitative data and link these to BCTs. The reviews of quantitative studies, by nature, were generally explicit in the reporting of constructs or strategies, but these were implicit in the qualitative review data. Given this, to identify which constructs or strategies may be related to patient adherence, a reviewer first familiarized herself by reading and rereading the extracted data. Inductive line-by-line coding was conducted, taking into consideration the meaning and context of different barriers and facilitators to adherence. The codes applied were the constructs and strategies identified from the quantitative findings or concepts from relevant psychological theory. The reviewer then followed the same process of mapping to BCTs as that used for the quantitative review data.

The initial findings and mapped BCTs from the quantitative and qualitative data were discussed within the research team to provide a sense check and to draw from additional expertise. In light of this, the first author further refined the associations between constructs and strategies and BCTs.

The electronic database searches identified 1701 references. After duplicates were removed, 1507 references remained. A reviewer screened all title or abstracts, and a second reviewer screened 29.86% (450/1507) of these (84% agreement). Moreover, 1296 references were excluded at the title-abstract stage. A reviewer read the full texts of the remaining 211 references, and the second reviewer read 29.9% (63/211) of these (71% agreement). Consensus on eligibility at both the title-abstract and full-text stage was reached through discussion. Any disagreements were resolved by discussing proposed reasons for inclusion and exclusion against the eligibility criteria. A total of 25 eligible systematic reviews (19 quantitative reviews, 3 qualitative reviews, and 3 mixed-methods reviews) published between 1995 and 2017 were included in this review—23 systematic reviews from database searches and 2 additional systematic reviews known to the researchers (Figure 1 illustrates the study inclusion process). From this point, 2 mixed-method reviews [33,34] have been integrated with the reviews of quantitative studies and the third mixed-method review [35] has been integrated with the reviews of qualitative studies.

In total, 14 theoretical constructs and 13 behavioral strategies were extracted from 18 of 21 systematic reviews of quantitative studies because they were related to medication adherence (Table 1). There was variation in the methods used by review authors to analyze the relationship of a construct or strategy to medication adherence. The different approaches to analysis included moderation analysis, correlation analysis, calculating a tally of the number of primary studies that had found a statistically significant effect when a particular construct or strategy was present and, where possible, associated effect sizes, and qualitative comparative analysis to determine the necessity or sufficiency of an individual construct or strategy to influence medication adherence.

The 14 theoretical constructs and 13 behavioral strategies were mapped to 34 BCTs (Table 2). Each construct or strategy was linked with between 1 and 13 associated BCTs.

In total, 17 theoretical constructs and 12 behavioral strategies were coded in the qualitative data [35,54-56] (Multimedia Appendix 2). Of these, 11 were not identified by the reviews of quantitative data, including 6 constructs: identity, motivation, negative emotions, response efficacy, social comparison, and social context and 6 strategies: credible source, demonstration of the behavior, problem solving, self-adjustment and experimentation with medication dose and frequency, self-management strategies, and self-monitoring of outcome(s) of behavior). The 17 constructs and 12 strategies were mapped to 46 BCTs; 12 BCTs were identified in addition to the 34 BCTs mapped from the quantitative review findings. Each construct or strategy had between 1 and 25 associated BCTs (Table 3).

In total, a pool of 46 BCTs were mapped from the theoretical constructs and behavioral strategies identified in the included reviews (Textbox 1, see Multimedia Appendix 3 for BCT definitions from the study by Michie et al [27]).

This rapid review identified 25 published systematic reviews of medication adherence in patients with physical health conditions and extracted a total of 20 theoretical constructs and 19 behavioral strategies associated with medication adherence. These constructs and strategies were mapped to 46 applicable BCTs from the BCTT version 1 [27], which can be used in future medication adherence interventions. In the first stage, the reviews of quantitative research gave rise to 34 BCTs related to adherence in a broad range of chronic physical health conditions. To ensure the specific context of adherence in type 2 diabetes was accounted for, 12 additional BCTs were identified following review of the diabetes-specific qualitative data.

This review has several strengths. It is the first rapid review of systematic reviews that aimed to identify theoretical constructs and behavioral strategies related to medication adherence in patients with type 2 diabetes. We have followed established and explicit systematic review methodology for this rapid review; thus, minimizing bias in literature searching, retrieval, and appraisal [57]. Furthermore, taking a rapid systematic approach to synthesize an evidence base and using systematic review methods to consider existing systematic reviews are 2 relatively new approaches to evidence synthesis. In addition, this review proposes a novel approach to develop message content, which could be used to inform future brief message–based or other electronic health and mobile health (mHealth) interventions. Use of the methodology of this review may lead to the identification of promising BCTs for promoting alternative health behaviors, as it is likely that different BCTs will be relevant to different contexts. This has produced findings in line with the validated BCTT [27], a leading classification system by behavior change researchers. Finally, the use of theory is often missing or inadequately reported in interventions for enhancing adherence in patients with type 2 diabetes, which makes it difficult to interpret why, how, and where theory may have impacted intervention success [26]. This rapid review starts to address these issues, proposing why, how, and where to incorporate BCTs in brief message intervention design.

This review has several limitations. We used a simple search strategy to facilitate a rapid review. Although this was appropriate for our aims, forward and backward citation searching may have improved the comprehensiveness of the literature identified. A number of systematic reviews were excluded because the independent relationship between theoretical constructs and behavioral strategies and medication adherence were not reported, and instead, the focus was on establishing the overall effectiveness of generally complex interventions. The quality of the included reviews was not assessed and, therefore, the possible risk of bias in the included systemic reviews is unknown. However, eligible reviews had to have been conducted systematically to maintain a minimum requirement of methodological rigor. Furthermore, consensus on the definition, and essential methodological processes, of rapid reviews has not been reached [58-60] and the necessity of the quality assessment stage has been debated [30]. We adhered to a set of defining characteristics proposed for rapid reviews by recent research [30]. In a modified Delphi consensus study, and based on the opinion of 66 literature experts, Kelly and colleagues [30] concluded that rapid reviews (1) are conducted in a shorter time frame than systematic reviews (completed in up to 3 months, compared with an average 15 months for a full systematic review) [61-63], (2) use the most systematic and rigorous methods to synthesize evidence and answer the research question(s) as the time limit permits, (3) tailor methods typical of a systematic review to accelerate the review process, and (4) are transparent in reporting all methods and findings. By following these guidelines, we have produced a methodologically strong rapid review.

Our method of mapping was adapted from the intervention mapping approach and incorporated the subjective views of the authors. In future, primary research may benefit from using more rigorous and reproducible methods of mapping. This may be achieved through the use of established coding frames and mapping techniques derived more heavily from the intervention mapping literature. Valuable research in the area of mapping theoretical constructs to BCTs is underway and this should serve to progress and inform future endeavors of a similar nature [64]. However, there is currently no consensus on how to do this [64] and, as such, our approach was appropriate for our aim to identify a selection of BCTs that can be used as a basis for future research.

The primary practical contribution of this review is the identification of promising intervention content in the form of a set of 46 BCTs. These could be used by researchers to form the basis for developing brief messages in interventions to promote medication adherence in people with type 2 diabetes. Such interventions are a promising avenue for improving adherence [26]. Given the aims of our wider program of research, we advocate this particular application of our findings, but our findings are not limited to brief messages or populations with type 2 diabetes. The comprehensive list of BCTs could be incorporated into other modes of intervention delivery, with a variety of interactive capabilities, for example, mobile phone apps. In addition, the 34 BCTs mapped solely from the quantitative review findings may be applied more broadly to medication adherence in chronic physical health conditions, provided unique social and contextual influences on adherence were taken into account.

The primary research implication of this review is that it facilitates the development of a body of brief messages based on the identified BCTs. Each message could incorporate a BCT and use it to frame and support different aspects of adherence behavior in people with type 2 diabetes. It will be important to ensure that messages accurately reflect the BCTs with which they are associated. The process of content development may be improved by seeking expert input. This is the next planned step in our program of research; health care professionals and behavioral scientists will be asked to generate brief messages based on the target BCTs. These messages and associated BCTs will be further tested in people with type 2 diabetes to assess credibility and acceptability, and subsequently in behavioral scientists to assess fidelity. Following this, a feasibility trial and a full RCT will assess the effectiveness of the brief message intervention to support medication adherence in people with type 2 diabetes. The approach of this review may be particularly suited to the area of adherence, given that the need for new approaches has been identified by leading systematic reviews [13]. More broadly, other researchers may wish to adapt our approach to develop novel interventions for other behaviors or populations where there is a need.

This review is a starting point; we have cast a wide net to suggest new intervention content, given that existing interventions have been found to have little effect on medication adherence. Further research is needed to establish more precisely which BCTs might be most effective at supporting medication adherence in particular contexts as well as the factors that may influence effectiveness. In the context of diabetes, these factors could include health literacy, diabetes health competency, and numeracy. This is complex to establish, and such questions were beyond the scope of this review. When selecting BCTs, researchers should be guided by what they understand of the context and mode of intervention delivery and nominate BCTs that suit their particular setting, population, and research aims while considering any relevant disease-specific research. It will be important to achieve a balance between learning from previous evidence in the field and testing novel approaches to support adherence.

This review has identified a range of theoretical constructs and behavioral strategies related to medication adherence and mapped these to 46 BCTs that may show promise in supporting adherence in people with type 2 diabetes. We propose developing and testing the effectiveness of brief message content in SMS text messages based on the 46 BCTs and have, thus, described a novel approach to designing the content of brief messages to optimize mHealth interventions.